← All Content & Research
Workforce & Market Brief · AI Photonics

NVIDIA bet $2B on a Texas fab. It still has to be staffed.

Coherent is quadrupling wafer capacity at its indium-phosphide (InP) photonics fab in Sherman, Texas. These are the lasers that move data between GPUs in every AI datacenter. The capital is committed and the market is exploding. What stands between the two is roughly 550 advanced-manufacturing and technical roles that have to be filled, in a Texoma labor pool Texas Instruments is also drawing on for a $30B megafab. This brief answers four questions: who can train the workforce, how fast the market is growing, what the product is worth, and what it costs if the people aren't there.

SECTOR InP Photonics
SITE Sherman, TX
HORIZON 2030 / 2033
PREPARED June 2026
The capital
$2B+
NVIDIA's March 2026 investment in Coherent, plus a multibillion-dollar purchase commitment and a $50M CHIPS letter of intent. The money doubles the Sherman fab's footprint and quadruples its wafer capacity.
The constraint
550+
Direct advanced-manufacturing, engineering, and technical roles that must be filled (1,000+ jobs total), in a town where TI is simultaneously hiring ~3,000 for its own megafab.
Solvable
The verdict
Trainable in weeks, if the pipeline starts now.

This is the hopeful counterpoint to the nuclear workforce story. A reactor operator takes three to five years to license. Photonics and semiconductor technicians can be trained in 10 days to two years. The gap is real and front-loaded, but it is solvable on fab-build timelines, if the apprenticeship, college-cleanroom, and shared-pipeline moves start now. The cautionary precedent is concrete: TSMC delayed its Arizona fab specifically because it couldn't find enough trained U.S. staff.

On track
Tight ◄ we are here
At risk
Idle fab
Capital committed · the fab is funded Start now → staffed on schedule · wait → idle capital
If the pipeline starts now

Staffed on schedule

Short-cycle training runs 10 days to two years. A program launched in 2026 graduates technicians as the expanded fab comes online. Capital meets crew, and the wafers ship.

If it doesn't

Idle capital

A finished fab produces nothing until it's staffed. TSMC delayed Arizona for exactly this reason. In Sherman, TI's megafab next door makes the competition for technicians worse, not better.

01

The deal, in one paragraph

Start with what was actually signed, because the numbers set the stakes for everything that follows.

What Coherent signed

Coherent signed a CHIPS Act letter of intent for up to $50M to expand its 6-inch indium-phosphide (InP) photonics fab in Sherman, TX, doubling manufacturing space and quadrupling wafer capacity. It builds on ~$20M from the Texas Semiconductor Innovation Fund and the Sherman EDC, and a $2B NVIDIA investment (March 2026) plus a multibillion-dollar purchase commitment. At completion: 1,000+ jobs, including 550+ direct advanced-manufacturing, engineering, and technical roles. InP photonics are the lasers that move data between GPUs in AI datacenters.

$50M
CHIPS Act letter of intent to expand the Sherman fab
Coherent, Jun 2026
$2B
NVIDIA investment, March 2026, plus a multibillion-dollar purchase commitment
NVIDIA
Wafer-capacity increase; manufacturing space doubled
Coherent
550+
Direct technical roles to fill, of 1,000+ total jobs
Coherent / NVIDIA
02

Who can train this workforce

Four tiers feed this fab: a local community-college anchor, a degreed-engineer pipeline, Texas's state-funded accelerators, and a thin national network of photonics-technician programs, the scarce specialty. The institutions exist. The InP-specific piece is what's missing.

InstitutionProgramTypeNote
Grayson College
Sherman / Denison, TX
Electrical Engineering Technology, semi equipment & automationAAS CertThe local anchor. Already names Coherent as an employer. Stackable: L1 cert ~6mo, both certs 12mo, AAS 18mo; ~$60k start. Won a $350k TWC grant (Oct 2025) to train process technicians.
UT DallasNorth Texas Semiconductor InstituteBSCo-leads a consortium with Grayson; $700k TSIF cleanroom grant. The degreed-engineer pipeline Coherent cites.
Texas A&MPhotonics / Electrical EngineeringBSStrong photonics and EE programs feeding North Texas.
TSTC
Texas State Technical College
Accelerated Semiconductor Technician Program10-week$3.5M TSIF grant. The fastest in-state on-ramp to a fab-floor role.
Austin CC + UT AustinTexas Institute for Electronics: Semiconductor Training CenterAAS CertState-backed training center built to the industry's workforce spec.
Temple CollegeStackable cert → AAS → BASAAS BASIncludes military-transition training, a ready source of disciplined recruits.
Indian Hills CC
Iowa
Laser / Electro-Optics TechnologyAAS750+ photonics-technician graduates placed, among the deepest benches in the country.
Indian River State College
Florida
Photonics (home of LASER-TEC)AAS CertCoordinates the national photonics-technician curriculum (OP-TEC / LASER-TEC).
Stonehill College
Massachusetts
Photonics certificate + paid internship9-mo CertA fast, work-integrated route into the specialty.
AIM Photonics Academy
MIT
InP / PIC hands-on bootcampsBootcampThe closest existing model for InP-specific upskilling: the module Sherman needs.
The gap to flag

Grayson's curriculum is semiconductor-equipment-generic, not InP/photonics-specific. The missing piece is an AIM-Photonics-style InP module layered on top, funded by the CHIPS and TSIF money already on the table. That single addition is the difference between a generic technician pipeline and one that can actually run an indium-phosphide fab.

Proven employer playbooks to copy

None of this is unprecedented. TSMC Arizona runs a registered technician apprenticeship. Micron + Onondaga CC built a $15M cleanroom stocked with real fab tools. Maricopa's 10-day Quick Start has certified 900+. The school-into-the-plant model already works; Sherman's job is to copy it fast.

03

How fast the market is growing

The fab isn't expanding on a hunch. InP photonics is the supply-constrained input to AI networking, and every credible forecast points the same direction: up and to the right, fast.

InP photonics market size
2024today
$5.8B
2033forecast
$28.4B
Today
2033 forecast · ~19% CAGR
MarketNowForecastGrowth
InP photonics (overall)$5.8B (2024)$28.4B (2033)~19% CAGR
Pluggable optics~$6B (2023)~$25B (2030)~4×
AI-datacenter opticsn/a~$100B/yr by 2030LightCounting
Optical transceivers (global)n/a~$73B by 2030Fuji Keizai
AI-datacenter InP demandn/an/a~85% CAGR, 2026–30
Co-packaged optics (CPO)$46M (2024)$8.1B (2030)~137% CAGR (Yole)*
~19%
CAGR for InP photonics overall, 2024–2033
Market aggregators
~$100B
AI-datacenter optics market per year by 2030
LightCounting
~85%
CAGR in AI-datacenter InP demand, 2026–2030
Industry est.
4–8
InP laser chips in every 800G optical module
LightCounting
The direct driver

Every 800G optical module uses 4–8 InP laser chips, and the industry is moving from 800G to 1.6T as standard in AI datacenters. That transition is the direct reason for Coherent's 4× capacity move.

* Estimates conflict widely on the newest segment: Yole pegs co-packaged optics at $8.1B by 2030; 360iResearch sees only $4.7B. Treat the direction as solid and any single point estimate as directional.

04

What the product is worth

InP photonics is a small line item that gates a very large one. That leverage is the whole reason a chipmaker put $2B into a single fab.

A small fab that gates a huge market

InP photonics is a high-value, supply-constrained input to the entire AI buildout. Most of the world's InP is still made on 3- and 4-inch wafers. Coherent's move to 6-inch quadruples usable area and cuts die cost by more than 60%, which is why NVIDIA invested $2B and signed a multibillion-dollar purchase commitment. In effect, a relatively small photonics fab gates a market measured in the tens of billions per year, sitting underneath NVIDIA's pledged ~$500B of U.S. AI infrastructure. The leverage ratio, dollars of AI infrastructure enabled per dollar of fab, is extreme.

The input

A supply-constrained bottleneck

Most InP still runs on 3-/4-inch wafers. The 6-inch jump quadruples usable area per wafer and cuts die cost sharply, turning a boutique process into a volume one.

>60% die-cost reduction
What it gates

The AI buildout downstream

InP lasers move the data between GPUs. The optics market they feed is measured in tens of billions a year, and it sits under the half-trillion in U.S. AI infrastructure NVIDIA has pledged.

~$500B U.S. AI infra pledged
05

The cost if the workforce isn't there

The risk is not abstract. It has already happened elsewhere, and Sherman's local conditions make it sharper than the national average.

Risk 1 · Idle capital

An unstaffed fab produces nothing

A new fab is a hundreds-of-millions-to-billions building that earns zero until it's staffed. The ~$650M Sherman expansion only pays off when the 550+ roles are filled.

~$650M expansion at stake
Risk 2 · Precedent

The TSMC Arizona delay

TSMC pushed back its Arizona fab specifically because it couldn't find enough trained U.S. staff. The workforce risk has already delayed a flagship fab once.

A delay, not a hypothetical
Risk 3 · National shortfall

67,000 short by 2030

SIA and Oxford Economics project the U.S. will be ~67,000 semiconductor workers short by 2030, or 58% of new jobs. Of those, ~26,400 are technicians, the single largest at-risk category.

~26,400 technicians short
Risk 4 · The town next door

TI's $30B megafab

Texas Instruments is building a $30B+ megafab (~3,000 jobs) in the same town, drawing from the same Texoma technician pool. Sherman's staffing risk is structurally higher than the national average.

TI hiring ~3,000 locally
The downstream bottleneck

Because InP optics gate AI-cluster networking, a photonics-technician shortfall doesn't just slow one fab. It throttles the broader AI-infrastructure buildout that depends on the supply. The constraint is small and specific, and that is exactly what makes it dangerous.

The hopeful counterpoint

Nuclear operators take three to five years to license. Photonics and semiconductor technicians can be trained in 10 days to two years. The gap is real and front-loaded, but it is solvable on fab-build timelines, if the apprenticeship, college-cleanroom, and shared-pipeline moves start now.

06

The bottom line

Where this leaves us

The capital is committed, the market is compounding at double digits, and the product gates the AI buildout. The only variable left is the crew. Unlike the nuclear workforce problem, the training here is fast. So the work is narrow and concrete: layer an InP/photonics module onto Grayson's existing pipeline with the CHIPS and TSIF money already pledged, copy the apprenticeship and cleanroom playbooks that already work, and start now, before TI absorbs the local technician pool.

The limit isn't money, demand, or technology. Those are all moving. The limit is 550 trained people, on a fab-build clock. That is an eminently winnable race, but only if it starts today.

Start now
The mandate
Capital's here. The market's here. Build the crew.

A program that stands up an InP-specific technician track in 2026 graduates its first cohorts as the expanded fab comes online, exactly when Coherent and NVIDIA need them. The institutions, the funding, and the proven playbooks all already exist. What's missing is the decision to assemble them before the local labor pool is spoken for.

Whoever stands up the InP module, copies the apprenticeship model, and starts before TI's megafab absorbs the Texoma technician pool will decide whether $2B of committed capital ships wafers or sits idle.

Caveats

"World's first 6-inch InP fab" is Coherent's own claim (NVIDIA repeats it as "what it calls"). Prior incentives are cited as ~$20M (Coherent) vs ~$17M (NVIDIA); expansion scale is ~$650M in most press, lower in narrower figures. Photonics-technician demand rests largely on one 2021 AIM/MIT study (~2,200 engineering-technicians/yr; ~20 vs ~140 programs). No BLS "photonics technician" line item exists. CPO market estimates disagree by an order of magnitude (Yole vs 360iResearch). Wages are BLS OEWS May 2024 medians, with "photonics tech" approximated by electro-mechanical / electronics-technician bands.

About Project Arklight

Project Arklight is a workforce-development company rebuilding how America trains skilled industrial labor.

We run a software-enabled trade school, Trade School 2.0, that assesses, trains, and deploys production-ready operators (electricians, machinists, welders, fabricators, and semiconductor/photonics technicians) to the companies reshoring American manufacturing and energy. We also publish original research on the skilled-labor gap: where it is, how deep it runs, and what it takes to close it. As the Sherman fab shows, the limit on the AI buildout isn't capital or demand. It's the supply of trained people. Project Arklight exists to remove it.

Trade School 2.0

We build the builders.
Build your workforce.

You can't scale what you can't staff. We assess, train, and deploy operators who ship to spec on day one, co-designed to your floor and paid on outcomes. Co-design your first cohort.